This model is an attempt to get Stable Diffusion to invent a new concept of some sort. To do this I picked 100 images from my file of training epoch samples, and on half of them I included the non-word "XQXQXQX" as a tag. The images were of widely differing styles, and I didn't consider the contents of any when deciding to include the tag. What would SDXL make of it? Is XQXQXQX a noun? A verb? An adjective? Apparently it depends on how you use it.This is meant to be played with and explored. It can be hard to predict, without being particularly unstable. Have fun! (And please post some of your results.)

The XQXQXQX Experiment

For this version I used the same dataset, except that I deleted the XQXQXQX tag and added a QXQXQXQ tag to the images that weren't tagged with XQXQXQX last time.

Files with SHA256: 222347830eba210504eb5b892066c79602d96fd6fce194b8e571bc330066358c